Static spatial Durbin panel model

The static SDM treats the spatial-specific effects as fixed effects and takes the spatial lag of both the dependent and explanatory variables. Observing the model diagnostics in Table 7.7, the SDM shows better statistics of model fit compared to the SAR. The R-squared of the SDM (12.4%) is greater than the SAR (4.5%); the residual variance (0.062) is smaller than the SAR (0.063); the log-likelihood (-297.76) is higher than the SAR (-368.56); and the AIC and BIC (619.52 and 695.35) are smaller than the SAR (751.13 and 795.36). These suggest that there is a better model fit when considering the spatial lag of both the dependent and explanatory variables.

Moving on, the estimations of the impact of spatial clustering and agglomeration economies on the labour productivity of T&H firms are discussed. First, the direct effects of the explanatory variables, i.e. the effect of the variables on labour productivity within a LAD, will be discussed. The degree of clustering of T&H firms is negatively associated with labour productivity within a LAD – a statistically significant 0.242% decrease in labour productivity for every one percent increase in the degree of clustering. As the degree of clustering of T&H firms increase within a spatial area, localised competition may increase, which may constitute a form of agglomeration diseconomies (McCann and Folta, 2009). As the overall number of T&H firms increases and they locate closer together, the performance of individual firms can decrease due to increased competition in the markets (Baum and Mezias, 1992). Competition for scarce local resources, i.e. labour and land, can increase the cost of firms in a local area, which further causes detrimental effects on productivity (Combes and Duranton, 2006). However, spatial clustering may not necessarily only imply competition. Firms that cluster together may also benefit from cooperative behaviour via agglomeration economies or positive

spatial spillover effects of knowledge, for example, which can generate greater efficiency and productivity (Viladecans-Marsal, 2004). Shaver and Flyer (2000) also argued that asymmetric

Table 7.7 Static spatial panel model coefficient estimations

Dependent variable: ln(T&H labour productivity)

Static SAR SDM

Direct effect

Degree of clustering ln(location quotient) -0.239

***

(-4.87)

-0.242*** (-4.92) Labour market

pooling ln(skilled labour pool)

0.087*** (6.68)

0.076*** (5.87)

ln(formal entry qual) -0.117

***

(-9.81)

-0.114*** (-9.66) Knowledge spillover ln(last job in T&H) 0.008

(0.85)

0.008 (0.86) Control variable non-T&H labour productivity 0.007

(0.97)

0.015* (1.91) Spatial spillover effect

Dependent variable W*ln(T&H labour productivity) 0.643

***

(54.33)

0.607*** (48.24)

Degree of clustering W*ln(location quotient) 0.086

(0.96) Labour market

pooling W*ln(skilled labour pool)

0.215***

(8.37)

W*ln(formal entry qual) 0.041

*

(1.72)

Knowledge spillover W*ln(last job in T&H) 0.045

**

(2.38) Control variable W*non-T&H labour productivity 0.006 (0.53) Model diagnostics Observations 4,103 (n=373;T=11) 4,103 (n=373;T=11) R-squared 0.045 0.124 Residual variance 0.063 0.062 Log-likelihood -368.56 -297.76 AIC 751.13 619.52 BIC 795.36 695.35 Hausman test 139.49*** _106.27***

Note: W=queen; *** _{indicates significant at the 0.01 level,} ** _{indicates significant at the 0.05 level,} * indicates significant at the 0.10 level

contributions to agglomeration can also generate mixed results in the agglomeration impacts on productivity, which requires further investigation. Thus, T&H firms may benefit from the agglomerative effects which occurs when firms cluster spatially (Rosenthal and Strange, 2003).Regarding the role of agglomeration economies in the labour productivity of T&H firms, the labour market pooling components show positive associations. A one percent increase in the skilled labour pool increases labour productivity by 0.076% and is statistically significant. This implies that there is a positive impact of a skilled labour pool in the LAD, measured by the share of T&H labour with National Vocational Qualifications (NVQs), or equivalent, on the labour productivity of T&H firms. A substantial literature has found a positive association between skilled labour pool and productivity because a denser human capital will enhance productivity (Rosenthal and Strange, 2001; Bathelt, 2008; Abel, Dey and Gabe, 2012) and because there is a higher possibility of better labour matches within a locality (Rosenthal and Strange, 2004). This can be evident in T&H according to the findings in Table 7.7.

In contrast, the variable ln(formal entry qual), which proxies the share of high skilled jobs in a LAD, is significantly negative (-0.114) on the labour productivity of T&H. This may be because high skilled jobs associate with high wages, which can put pressures on productivity due to the high costs related to production (McCann and Folta, 2009). Two variables are used to proxy labour market pooling and the key difference is that the ln(skilled labour pool) variable attempts to proxy the share of human capital, and ln(formal entry qual) variable attempts proxy the share of high-skilled jobs. The results may imply that having a pool of skilled labour, which is diverse in terms of level of qualifications, can improve T&H labour productivity, but a high costs related to high-skilled jobs can actually have negative implications on productivity.

The direct effect of knowledge spillovers on the labour productivity of the T&H firms within a LAD is shown to be positive but statistically insignificant. A one percent increase in the share of T&H employees who had a previous job in T&H will increase labour productivity by 0.008%. Yet, there is no statistical significance, which means that the effects are unknown. Researchers have argued that knowledge spillovers are not uni-directional – there can be inflows and outflows (Mariotti, Piscitello and Elia, 2010). When firms are in spatial proximity, it does not mean that firms will necessarily interact with each other and also interaction does not always mean positive spillovers (ibid.). Firms may absorb certain knowledge, but also lose it, and the net balance is not always positive. Lastly, the control variable, non-T&H labour productivity, which controls for the local effect, is positive and statistically significant,

meaning that the change in T&H labour productivity within a LAD is not sector-specific but moves in line with the rest of the local economy. This supports the descriptive statistics (Figure 6.1 in Chapter 6), which shows a similar trend between the average labour productivity of T&H and non-T&H sectors across the years from 2006 and 2016.

The spatially lagged variables (expressed as W*variable name) estimate the indirect spatial spillover effects of agglomeration economies on labour productivity, but also the spatial spillover effects of labour productivity from one LAD to another. Since the model uses the queen contiguity spatial weights matrix (W=queen), neighbouring LADs are defined as LADs that share their borders (including the sides and edges of the shape of the LAD) with the focal LAD. First, the spatial spillover effect of labour productivity shows that if there is a one percent increase in T&H labour productivity in the neighbouring LADs, then the labour productivity increases by 0.607% in the focal LAD; this is statistically significant. Regions with similar visitor economies can attract a similar visitor market, and coupled with spatial proximity between the regions, this can generate significant productivity spillover effects across the regions (Yang and Wong, 2012). Additionally, neighbouring regions can learn from demonstration or imitation, where T&H firms can learn from other firms in highly productive regions (Hall and Williams, 2008). This also associates with knowledge spillovers which can induce productivity spillover effects across neighbouring regions (Capello, 2009). Additionally, the degree of clustering of T&H firms in a LAD is positive when considering its spillover effect (0.086% increase in labour productivity) but is statistically insignificant. It is important to acknowledge that markets can extend across more than one LAD, given LADs are administrative units. Yet, they have been considered as a reasonable approximations of local labour market areas according to UKCES (2014). Given the negative direct effect of the degree of clustering on T&H labour productivity within a LAD, localised market competition across LADs could also be anticipated across geographical boundaries.

Regarding the spatial spillover effects of agglomeration economies, a one percent increase in the skilled labour pool will increase labour productivity by 0.215%, which is statistically significant. This suggests significant complementary effects of a skilled labour pool across neighbouring regions, which can boost T&H labour productivity. This is consistent with previous empirical studies, such as Ramos, Suriñach and Artís (2010), Melo and Graham (2014) and Wixe (2015). Human capital can spillover from one region to another and influence the neighbouring regions. Unlike the direct effect, the spatial spillover effect of ln(formal entry

qual) is positive (0.041%) when considering its effect across neighbouring regions. Thus, the spatial spillover effect of labour market pooling significantly contributes to T&H labour productivity growth across neighbouring LADs. Potential movements of labour across localities or labour markets in close proximity (Di Addario and Patacchini, 2008; Wixe, 2015), whether that is daily commuting or economic migration, can improve labour productivity. This could be because new knowledge is brought into the local area via such movement, but also as new people can participate in the local labour market in different ways, which can potentially address the issue of skills mismatch between the local regions (López-Bazo, Vayá and Artís, 2004; Fingleton and López-Bazo, 2006; Rosenthal and Strange, 2008). Complementary effects of local labour markets and productivity between local regions can be generated through such spillover effects.

The spatial spillover effects of knowledge are also shown to have complementary effects of 0.045% across neighbouring LADs, unlike the direct effect which only considers the effects within a LAD. Based on the measure of knowledge spillover, existing literature contends that knowledge spillovers tend to be highly tacit in T&H, and that much of the knowledge is job- related and highly relevant with individuals drawing insights from previous work experiences (Yang, 2007, 2008, 2010). Paci and Usai (1999) asserted that knowledge spillovers are not locally bounded but can freely move across borders, in which spatial proximity can help firms to process knowledge sharing and diffusion across borders, which suggests the importance of agglomeration economies and their effects across spatial boundaries (Coe and Helpman, 1995). Skilled labour from regions of higher-level productivity can also carry knowledge and skills into new regions, which can benefit the productivity enhancement of the new region (Yang and Wong, 2012).

Positive human capital (knowledge) externalities have been confirmed by existing studies, suggesting that human capital in one region can also influence the neighbouring regions, implying regional spillover effects (Audretsch and Feldman, 2004; Møen, 2005; Ramos, Suriñach and Artís, 2010; Huang and Zhang, 2017). Knowledge can spillover locally and globally, which also influences the degree of impact on productivity. Bathelt, Malmberg and Maskell (2004) studied the difference between intra- (within a cluster) and extra- (between clusters) cluster knowledge. A local buzz (intra-cluster) allows opportunities for a variety of spontaneous situations where firms can interact closely to tackle them, and global pipelines (inter-cluster) allow individual firms to establish knowledge-enhancing relations to firms

external to the local milieu, and knowledge can spillover to other firms in extra-cluster through a local buzz (Greunz, 2003; Jacob and Groizard, 2007; LeSage and Pace, 2009). This could explain the significant positive impact of knowledge spillovers between LADs.

Amongst the SDM estimations, there is a contrasting effect between the direct and spatial effect. Regarding the direct effect of ln(formal entry qual) variable of -0.114, an increase in the share of high skilled jobs within a LAD will reduce the labour productivity of T&H firms within the same locality. In contrast, in terms of its lagged term, i.e. spatial spillover effects of ln(formal entry qual) is 0.041 and statistically significant. Thus, when we consider the spatial effects, it is complementary in productivity enhancement amongst the neighbouring LADs via a spatial feedback loop. A spatial feedback loop was introduced by LeSage and Pace (2009) where a one region impacts another and vice a versa, and even further than the next neighbouring regions and back to the focal region. It measures the average effect of the change in the explanatory variable on the dependent variable from the focal region to the neighbouring regions and back. These marginal effects are important because the direct coefficients of spatial models are often interpreted incorrectly as if they are simple partial derivatives (Golgher and Voss, 2016) – refer to Chapter 5 (section 5.2.1.2 and 5.5.1). Henceforth, when interpreting the coefficients of the SDM, theses marginal effects are crucial to interpreting the direct and indirect (spatial spillover) effects of the variables correctly.

Table 7.8 presents the impact measures of the static FE SAR and SDM. For the static models, only the long-run impact measures are estimated; the long-run effects imply the effects on Y at time T, as it goes to infinity, of a change in X (Doran and Fingleton, 2018). For the dynamic models, both the long-run and short-run impact measures are estimated (section 7.5.2). The direct effect measures the average effect of the change in explanatory variable (X) on the dependent variable (Y), including the feedback via the neighbouring LAD and back to the focal LAD. The direct effects in Table 7.8 are larger than the direct effect coefficient estimates presented in Table 7.7, which means that positive spatial feedback effects exist from passing through the neighbouring LADs and back to the focal LAD. The direct effects of the variables are similar to the coefficient estimates, but the direct effect of knowledge spillovers is statistically significant under the impact measures. This means that an increase in the share of employees who have had a previous job in T&H in the neighbouring LADs increases the T&H labour productivity in the focal LAD. Given the coefficient estimates of the spatial model tend to be incorrectly interpreted (Golgher and Voss, 2016), it can be argued that the direct effect

measure under the impact measures is more accurate. Thus, this supports the positive effects of intra-cluster knowledge exchange on firm innovation (Bathelt, Malmberg and Maskell, 2004), which can ultimately influence T&H labour productivity.

Table 7.8 Impact measures: direct, indirect and total effects (static)

Static SAR SDM

Long-run Direct effect

ln(location quotient) -0.269*** 0.251***

ln(skilled labour pool) 0.097*** 0.123***

ln(formal entry qual) -0.131*** -0.118***

ln(last job in T&H) 0.009 0.017*

non-T&H labour productivity 0.008 0.017***

Indirect effect

ln(location quotient) -0.398*** -0.131

ln(skilled labour pool) 0.144*** _0.615***

ln(formal entry qual) -0.194*** -0.066

ln(last job in T&H) 0.013 0.121***

non-T&H labour productivity 0.011 0.036

Total effect

ln(location quotient) -0.667*** -0.382

ln(skilled labour pool) 0.241*** 0.738***

ln(formal entry qual) -0.324*** -0.184***

ln(last job in T&H) 0.022 0.138***

non-T&H labour productivity 0.019 0.054**

Note: W=queen; *** _{indicates significant at the 0.01 level,} ** _{indicates significant at the 0.05 level,} * indicates significant at the 0.10 level

Source: ONS (2012, 2018a) and ONS. Social Survey Division (2018)

The indirect effect measures the average effect of the change in X of the focal LAD on the Y of neighbouring LAD or X of neighbouring LAD on the Y of the focal LAD. The indirect effects are only statistically significant for the variables ln(skilled labour pool) and ln(last job in T&H). These effects are greater in magnitude compared to the coefficient estimates in Table 7.7, but here the effect of ln(formal entry qual) on T&H labour productivity is statistically insignificant. Yet, the spatial spillover (indirect) effects of agglomeration economies (both elements of labour market pooling and knowledge spillovers) are significantly present on the labour productivity of T&H firms. This can help enhance labour allocation and matching across a wider geographically boundary, and subsequent labour movement can enact significant knowledge spillovers across these boundaries. This can suggest potential regional productivity growth via agglomeration economies in the T&H industry.

The total effect is the sum of the direct and indirect effect, which therefore measures the average effect of the change in X of the focal LAD on the Y of all the focal and neighbouring LAD. The effects are statistically significant except for the degree of clustering. Despite the indirect effect of ln(formal entry qual) being statistically insignificant, the total effect shows significance. This may suggest both complementary and competition effects of labour market pooling across neighbouring LADs. The high costs related to high-skilled jobs and the high labour mismatch, which T&H firms struggle to tackle, can be observed here (McCann and Folta, 2009). Nevertheless, the share of skilled labour (proxied by NVQs) show a larger positive effect on T&H labour productivity.

The key outcomes of the static SDM estimations are that the increase in the degree of clustering of T&H firms has had negative effects on labour productivity of T&H firms within a LAD but also across neighbouring LADs and back. This is possibly due to the fierce competition related to spatial proximity and increase in rivalry as the clustering increases in density. Labour market pooling variables have shown some partial positive effects regarding the pool of skilled labour but the effects on productivity where negative in terms of the share in high-skilled jobs. The high costs related to high-skilled jobs and the difficulties with recruitment and retention may reflect this negative impact. However, when considering the spatial spillover effects of labour market pooling across neighbouring LADs, they are significantly positive, implying the complementary effects of labour market pooling on productivity of T&H across regions. Knowledge spillover effects on labour productivity of T&H firms within a LAD were insignificant, but when considering its spatial spillover effects, the effects were complementary, suggesting the importance of human capital externalities and their spillover effects across geographical boundaries. Yet, based on the marginal effects, the direct, indirect and total effect of knowledge spillovers were significantly positive on T&H labour productivity.